The Pricing Paradox in AI Applications: From SaaS to Outcome-Based Pricing 🚀
In a recent post on X (formerly Twitter), Jason Liu shared a deep reflection on how AI applications are priced. At an office hours session focused on AI value capture, he arrived at an insight: existing SaaS (Software as a Service) pricing models are no longer sufficient, and a new approach is needed.
Current AI Pricing Models and Their Limits
Jason points out that most companies today are stuck in two traditional models:
- Usage-based:
- Charging based on actual consumption — tokens used, API calls made, and so on.
- Seat-based:
- Charging per user or per month.
The problem with both models is that neither is directly tied to the business value AI actually creates.
"We're still pricing AI like traditional SaaS. But it's time to price on outcomes."
What Is Outcome-Based Pricing? 💡
Jason argues that outcome-based pricing can reflect the true value of AI. He uses customer support automation as an example:
- Traditional model:
- Pay based on tokens or API calls
- Outcome-based model:
- "Pay $1 per resolved ticket"
- "Pay only for qualified leads that actually converted"
"What if you paid $1 per ticket resolved in customer support? Or only paid for leads that actually converted in sales outreach?"
This model means businesses only pay in proportion to the value AI actually delivers — a more rational arrangement from the buyer's perspective.
A Conceptual Challenge, Not a Technical One
Jason emphasizes that this shift is not a technical problem — it's a conceptual one.
"We're building AI that makes humans 2% more productive. But we should be building AI that can replace entire workflow chains."
The point is not simply to make people marginally more efficient, but to enable AI to fundamentally transform entire business processes.
The Virtuous Cycle of AI and Data
Jason envisions a future where AI applications generate data like sensors, and that data feeds back into building better AI models — a virtuous cycle:
- Every AI application generates valuable data
- That data leads to better models
- Ultimately, the most valuable dataset will be data about how humans interact with AI
Whoever Owns the Value Chain Wins
Jason is unequivocal: the companies that own the entire value chain will win.
"The biggest insight is that whoever owns the full value chain will win — not because they have a better model, but because they can price on outcomes."
He gives the example of voice AI: rather than charging per minute, a more forward-looking approach is to take a cut of the revenue the AI actually generates for a mechanic through bookings.
"Why build a voice AI that charges per minute? You could build a system that takes a percentage of the revenue generated for a mechanic through bookings."
The Pricing Paradigm Shift Ahead 🌟
Jason closes by stressing the need to move the fundamental question from "How much does it cost to run this service?" to "How much value does this service create?"
"The future belongs to those who can shift from 'how much does it cost to run this?' to 'how much value can this create?'"
A Range of Reactions in the Comments
The replies to Jason's post surfaced several perspectives:
- Questions from people wanting to join the office hours
- Concerns that AI may become so efficient that many users won't be willing to pay much beyond inference costs
- The analogy: "It's like pricing a shovel by the value of what you dig up — great for sellers, bad for buyers"
- A counterpoint: "That's just value-based pricing — it's compatible with usage-based pricing. It's really about inputs vs. outputs; usage-based models can focus on output metrics too"
Key Terms
- AI application pricing
- SaaS (Software as a Service)
- Usage-based / Seat-based pricing
- Outcome-based pricing
- Value chain
- Data virtuous cycle
- Business value creation
Jason Liu's post offers deep insight into how AI service pricing must evolve going forward. It's a reminder that innovation in business models and mindset matters just as much as technological progress. 💡